Split-sample Test Research Articles

Improved streamflow forecasting using self-organizing radial basis function artificial neural networks

Streamflow forecasting has always been a challenging task for water resources engineers and managers and a major component of water resources system control. In this study, we explore the applicability of a Self Organizing Radial Basis (SORB) function to one-step ahead forecasting of daily streamflow. SORB uses a Gaussian Radial Basis Function architecture in conjunction with the Self-Organizing Feature Map (SOFM) used in data classification. SORB outperforms the two other ANN algorithms, the well known Multi-layer Feedforward Network (MFN) and Self-Organizing Linear Output map (SOLO) neural network for simulation of daily streamflow in the semi-arid Salt River basin. The applicability of the linear regression model was also investigated and concluded that the regression model is not reliable for this study. To generalize the model and derive a robust parameter set, cross-validation is applied and its outcome is compared with the split sample test. Cross-validation justifies the validity of the nonlinear relationship set up between input and output data.

Investigating Insensitivity to Scope: A Split-Sample Test of Perceived Scheme Realism

<i>This paper considers the use of contingent valuation to estimate non-market benefits from separately valued nested goods. Survey respondents were asked referendum WTP questions regarding either a scheme to protect the “whole” area of a wetland or a scheme to protect some nested “part” sub-area. The survey design permitted firstresponse testing and comparison of part and whole values revealed mixed evidence of scope sensitivity. However, allowing for variation in the degree to which differing schemes are considered to be realistic revealed highly significant scope sensitivity. These results illustrate the need for a detailed understanding of the preferences and beliefs of respondents when performing scope sensitivity tests.</i> (JEL Q26)

Evaluation of hydrological model parameter transferability for simulating the impact of land use on catchment hydrology

During the past decades, the use of hydrological models for predicting the impact of land use on catchment hydrology increased considerably. The performance of those models is often judged by a simple split-sample test using historical discharge series. The derived parameter values are then assumed to be identical for the new land use scenario, apart from the crop and management characteristics that are adapted to the land use under study. This paper checks the validity of this assumption in an indirect way, by evaluating the transferability of the main controlling parameters of the semi-distributed SWAT model in a stepwise fashion: within the catchment, a neighbouring catchment and a catchment under a different environmental setting. The results indicate that there is a decline in model performance when parameters are transferred in time and space. Transfer within the catchment and to a neighbouring catchment gives for the case study still a reasonable performance, yet one should be careful when exchanging parameter values between regions with a different topography, soil and land use. These factors might influence the infiltration and percolation of water and so affect the associated model parameters.

Reliability of Model Predictions Outside Calibration Conditions

Predictions of probabilities and magnitudes of extreme events are essential for water management. One approach for flood estimation is the use of conceptual runoff models. This approach, however, can be questioned for the same reason as the approach of extreme-value statistics: the model has to be used for conditions far beyond those used for model development and calibration. In this study the HBV model, a conceptual runoff model, was applied to four different catchments and differential split-sample testing (calibration on years with lower runoff peaks and testing it on years with higher peaks) was used to evaluate model performance for the situation when the model has to be used to simulate runoff during conditions different from those observed during calibration. To assess the value of improved calibration different goodness-of-fit measures were used, which allowed to explicitly consider the ability of the model to simulate groundwater-levels and peak flows. The results indicated that applying a model to conditions different from those during the calibration period might not give accurate results and that improved calibration procedures might not automatically provide more accurate flood estimations.

Clinical Performance of an In-Line, ex Vivo Point-of-Care Monitor: A Multicenter Study

The management of critically ill infants and neonates includes frequent determination of arterial blood gas, electrolyte, and hematocrit values. An objective of attached point-of-care patient monitoring is to provide clinically relevant data without the adverse consequences associated with serial phlebotomy. We prospectively determined the mean difference (and SD of the difference) from laboratory methods of an in-line, ex vivo monitor, the VIA LVM Blood Gas and Chemistry Monitoring System (VIA LVM Monitor; Metracor Technologies, Inc.), in 100 critically ill neonates and infants at seven children's hospitals. In doing so, we examined monitor stability with continuous use. In vivo patient test results from laboratory benchtop analyzers were compared with those from the VIA LVM Monitor on paired samples. In a separate in vitro comparison, benchtop analyzer and monitor test results were compared on whole-blood split samples. A total of 1414 concurrent, paired-sample measurements were obtained. The mean differences (SD of differences) from laboratory methods and r values for the combined data for the VIA LVM Monitor from the seven sites were 0.001 (0.026) and 0.97 for pH, 0.7 (3.6) mmHg and 0.94 for PCO(2), 4.2 (9.6) mmHg and 0.98 for PO(2), 0.0 (2.9) mmol/L and 0.87 for sodium, 0.1 (0.2) mmol/L and 0.96 for potassium, and 0.3% (2.9%) and 0.90 for hematocrit. Performance results were similar among the study sites with increasing time of monitor use and between in vivo paired-sample and in vitro split-sample test results. The VIA LVM Monitor can be used to assess critically ill neonates and infants.

DEVELOPING PROTOCOLS FOR MEASURING BIOSOLIDS STABILITY

The general goal of this project is to validate test protocols that are commonly used to assess the stability of various biosolids products, and to specify a standard for each test method. Toward this end, the project team conducted a literature review, surveyed selected practicing facilities, and conducted a sampling and analysis program. Information concerning stability-testing protocols and data for each biosolids-stabilization technology and testing method reviewed were collected and evaluated. The biosolids-stabilization technologies reviewed were aerobic and anaerobic digestion, alkaline stabilization, and composting; the testing methods reviewed were specific oxygen uptake rate, volatile solids reduction, additional volatile solids reduction, pH and changes in pH, and carbon dioxide evolution. The report recommends specific protocols for each testing method, and presents intrinsic precision data for each protocol. In addition, data is presented from facilities using each protocol, to serve as a basis for evaluating the collective degree of variability associated with each stabilization process. Collective variability results from sampling, feedstock, the stabilization process, and the intrinsic variability of each testing method. Using split sample testing, interlaboratory data (variability among different laboratories) are presented. Issues associated with variability, error sources, shortcomings, and numerical criteria values are discussed for each method. This title belongs to WERF Research Report Series ISBN: 9781843396550 (eBook)

Statistically Based Methods for Verification Testing

State highway agencies currently use split-sample and independent-sample verification procedures for the purpose of comparing test results or verifying the contractor’s overall construction process in independent assurance (IA) programs and quality assurance (QA) construction specifications. Sampling procedures are reviewed, and new statistically based methods for developing comparison procedures in both IA programs and QA specifications are presented. The statistically based approach determines an appropriate mean difference between either split-sample test results or independent-sample verification results by recognizing sample size, field variability, and inherent risks. A practical application is given using collected field data from hot-mix asphalt (HMA) construction projects. State highway agencies and contractors will benefit from an understanding of the statistical concepts involved with split-sample and independent-sample comparison procedures and of what information an agency needs to determine if a contractor’s overall test results are accurate. The concepts and procedures presented can be applied to all areas of highway construction involving split-sample and independent-sample verification tests, such as base materials, concrete structures and pavements, and HMA pavements.

Application of a distributed physically-based hydrological model to a medium size catchment

Abstract. Physically based distributed models are rarely calibrated and validated thoroughly because of lack of data. In practice, validation is limited to comparison of simulated and predicted discharges in a catchment, or of simulated and observed piezometric levels in some calibrated wells. Rarely, internal noncalibrated wells or discharge stations are included in model evaluation. In this study, the fully distributed physically based MIKE SHE model was applied to the 600-km2 catchment of the Grote and the Kleine Gete, Belgium. Firstly, the MIKE SHE model was calibrated against both daily discharge measurements and observed water levels and then validated using a simple split-sample test. The observed discharges were simulated successfully in both the calibration and the validation period, while results for the piezometric levels differed considerably among the wells. In addition, a multi-site validation test for 2 internal discharge stations and 6 observation wells showed inferior results for the discharge stations and comparable results for the water table wells. As in the calibration and the split-sample test validation, water table fluctuations were predicted well in some wells, but with little agreement in others. This may be due to scale effects and to the poor quality of the data in certain areas of the catchment. Mainly, the lack of data made it difficult to simulate time series of internal catchment variables with acceptable accuracy so that even the calibrated and validated model could not provide reliable predictions of the water table over the entire catchment. Keywords: integral hydrological modelling; distributed code; MIKE-SHE; model performance; model calibration; model validation

Operational testing of a water balance model for predicting climate change impacts

The ability of water balance models to incorporate month–month or seasonal variations in climate, snowfall and snowmelt, groundwater fluctuations, soil moisture characteristics, and natural climatic variability makes them especially attractive for water resources studies of climatic changes. The use of conceptual models to explore the impact of climate changes has increased in recent years. Because of the success claimed for these studies, it is likely that computer simulation of catchments will increasingly be used by and for water resource managers as an aid to decision-making. There is therefore a need for a generally accepted method for demonstrating a model’s fitness for such use. The simple split-sample test method may be reasonable in the simplest case of the ‘filling-in missing data’ problem but certainly not if the express purpose of the model is to simulate records for conditions different from those corresponding to the calibration record, such as the problem with predicting the effects of climate changes where the data on the changed system are not (and cannot be) available for comparison with the model predictions. Thus, model validation must demonstrate ‘fitness for the said purpose’.In this paper, available model validation methods are reviewed and the hierarchical scheme for systematic testing of hydrological simulation models according to the modelling tasks, proposed by Klemes (1986) (Klemes, V., 1986. Hydrol. Sci. J. 31, 13–24) is discussed in detail and exemplified using the NOPEX water balance model (Xu, C.-Y., Seibert, J., Halldin, S., 1996. J. Hydrol. 180, 211–236) and NOPEX data.

Hierarchical testing of three rainfall–runoff models in small forested catchments

This study examined whether the quasi-distributed response function used in TOPMODEL provides superior performance for event simulation in small, temperate forested catchments, compared to lumped reservoir representations of runoff routing similar to those employed in many catchment hydrology models. The alternatives were a two-reservoir black-box model and a three-reservoir model structured to represent our perceptual model of runoff processes based on field observations. A second objective was to test the statistical significance of differences in model performance using a new approach that combines the Jackknife with analysis of variance (ANOVA). The models were tested against streamflow data from two small forested catchments using Klemes’ hierarchical scheme for operational validation. At levels 1 (split-sample test) and 2 (proxy-basin test) there were no statistically significant differences in model performance when expressed using the entire hydrographs. However, TOPMODEL did appear to provide superior fits to the peak flows. At level 3 (differential split-sample test), TOPMODEL performed statistically significantly better than both lumped models in both catchments. The statistical analysis for level 4 (proxy-basin differential split-sample test) indicates that TOPMODEL performed significantly better than the lumped models at one catchment but not at the other. The statistical design combining ANOVA with the Jackknife shows promise as a workable method for establishing statistical significance of differences in model performance indices based on the entire hydrograph. However, performance indices based only on the peak flows exhibited extreme skew that was not amenable to normalization by transformation. Further research should investigate alternative, robust methods for assessing statistical significance of model performance indices.

Estimation of atmospheric transmittance from upper-air humidity

A very simple model of atmospheric transmittance uses as its input regularly observed and archived measures of upper air humidity. The five model parameters were determined empirically to optimize the agreement with radiation measured at six sites in the northwestern contiguous U.S. in summer and winter 1990. The model uses the relative humidity at 500 hPa, the relative humidity at either 950 hPa or 800 hPa depending on the altitude of the station, and the precipitable water. Although the model does not use cloud observations explicitly, it is partitioned into a low-humidity stage and a high-humidity stage with markedly different dependences on relative humidity. In the low-humidity stage the transmittance is approximated as the product of factors for dry air and for water vapor. In the high-humidity stage the transmittance is the product of the transmittances of each of the two layers used; the product is a strongly decreasing function of relative humidity, as a representation of the effect of cloud. A split-sample test using measurements for spring and autumn indicated that the model parameters were not badly distorted by conditions unique to the summer and winter seasons. In terms of the average daily global shortwave radiation received at the surface, the rms errors are 31 W m −2 in spring, 33 in summer, 30 in autumn, and 18 in winter. The corresponding relative errors are 0.25, 0.17, 0.20, and 0.31. Because of high-frequency temporal variation in the model residuals, the error declines to about half of the daily error for 7-day averages and to about a quarter of the daily error for 30-day averages.

A prognostic model for the prediction of survival in cystic fibrosis.

BACKGROUND: The treatment for endstage cystic fibrosis is, where appropriate, double-lung, heart-lung or, occasionally, heart-lung-liver transplantation. Optimising the timing of transplantation depends upon an accurate prediction of survival, but while...

Precision, Accuracy, and Managed Care Implications of a Hand-Held Whole Blood Analyzer in the Prehospital Setting

Hand-held portable clinical analyzers permit the rapid measurement of whole blood electrolytes, glucose, blood urea nitrogen (BUN) and hematocrit. Knowledge of these values in the field might aid radio telemetry emergency department physicians in the field treatment and triage of patients. The purpose of this study was determine if the analyzer could function in the hostile prehospital environment. In phase 1, analyses of control electrolyte (n = 30) and hematocrit (n = 28) solutions were performed in a moving ambulance by paramedics to determine precision performance. The F-statistic was used to compare variances against reference values and no significant differences were found. In phase 2, prospective split-sample testing of 57 whole blood samples drawn in the field were analyzed on 2 machines by paramedics in a moving ambulance, and then again within 10 minutes of arrival at the receiving hospital emergency department. Regression analysis between ambulance and emergency department venues revealed high correlation (r) values: sodium (Na)+ (0.93), potassium (K)+ (0.99), chloride (Cl)- (0.89), BUN (0.99), glucose (0.99), hematocrit (0.95), and hemoglobin (0.92). A hand-held whole blood analyzer can be reliably used in the field to obtain blood chemistry and hematocrit values. There was excellent correlation between field and hospital emergency department values. Clinical pathologists extend their oversight and consider encouraging emergency physicians to obtain field blood chemistry values in research studies aimed at improving medical treatment and patient triage in the prehospital setting. We speculate that these results might be important to managed care groups because knowledge of blood chemistry values in the field might provide physicians with objective, criteria-based data on which to triage patients to the emergency department, to an ambulatory care setting, or to a community health center with attendant cost savings.

Modelling yield changes after strip thinning in a mountain ash catchment: an exercise in catchment model validation

Lumped catchment models spatially average hydrologic processes at the hillslope scale or larger. Typically they can only be calibrated to streamflow data, and judgments on the reliability of their predictions must be based upon the results of a split-sample test in which part of the streamflow record is used for calibration and the remainder reserved for an independent test of the model. However, it has been argued that such a split-sample test is not an adequate test of model structure or of the hypotheses upon which the model is based because the independent record is typically hydrologically similar to the calibration record, rendering the test an exercise in model interpolation. A stronger test of model structure would force the model to extrapolate beyond the range of conditions encountered in the calibration record. One way to achieve this is to select a catchment which undergoes a major land use change during the independent test period. The small-scale experiments of forest hydrology are particularly important in this regard because they typically provide high-quality data on the effects of radical land use changes. This study illustrates application of this modified split-sample approach for the Crotty Creek experimental catchment which underwent a 6 year strip thinning treatment that removed 42% of its forest cover. Two lumped catchment models, the SDI and CATPRO models, were compared. To make use of the full rainfall and streamflow record it was necessary to modify these models to differentiate between the interception and evapotranspiration fluxes of the forest and the scrub communities which developed in the cleared strips. Both models were calibrated to the pre-treatment record during which there was 100% forest cover. Only the scrub parameters were calibrated to the post-treatment record. It was found that the SDI model, even though it had been previously extensively tested, failed to predict adequately the high monthly flows that followed the strip thinning treatment, whereas the CATPRO model, using a different conceptualization of quickflow, provided a reasonably good description of monthly water yields before and after the strip thinning treatment.

Unit hydrograph estimation with multiple events and prior information: II. Evaluation of the method

Abstract The method of regularization with additional information proposed by the authors is evaluated in a split sample test which compares it with the least squares and ordinary regularization methods. Data from 30 catchments are used. Additional information is provided by a simple conceptual model either fitted by a regression on physical characteristics of the catchment or fitted to the catchment lag determined from the data. The method is very useful when only a few rainfall/runoff events are available for analysis. When a large number of events is available, the ordinary regularization method is sufficient.

Implementation of capillary blood glucose monitoring in a teaching hospital and determination of program requirements to maintain quality testing

purpose: To study the implementation of bedside capillary glucose monitoring using a hospital-wide quality control (QC) program. methods: A prospective study of QC performance in 7 outpatient and 39 inpatient treatment units was performed in a large teaching hospital over a 2-year period. Approximately 800 nurses were trained to perform bedside capillary glucose monitoring (Accu-Chek II, Boehringer-Mannheim, Indianapolis, IN). An eight-point QC program was instituted including proficiency testing, instrument maintenance, performance of daily controls, storage of reagent strips and supplies, instrument calibration, and documentation procedures. results: Comparison of laboratory and bedside test results (split-sample proficiency testing) revealed Y = 1.004X + 7.26, r = 0.95, with a mean percent difference of −4.2% (p <0.001). Less than 7% of results fell outside ± 20% of the laboratory results. QC scores (0 = worst to 4 = best), based on adherence to the QC program, improved from 0 on the first inspection to 3.7 ± 0.17 by the 11th inspection. The most common QC deficiencies were failure to perform split-sample testing (41.4%) and failure to perform instrument maintenance (30.2%). Significant differences were noted in the QC performance of different types of medical services. During the 2-year study period, the total number of glucose assays performed in the clinical laboratories decreased by 22.2% concurrent with initiation of bedside testing. The number of instruments in the hospital increased from 10 to 46. conclusions: Bedside capillary glucose assays can be widely implemented in large hospitals with an acceptable degree of accuracy. QC programs with frequent inspections are necessary to identify units that function inadequately, and a formal disciplinary policy is required to ensure compliance with the program.

Certification of Cholesterol Measurements by the National Reference Method Laboratory Network with Routine Clinical Specimens: Effects of Network Laboratory Bias and Imprecision

The National Reference Method Laboratory Network has initiated a program to certify clinical laboratory cholesterol measurement performance by using routine clinical specimens. Clinical laboratory and reference laboratory measurements of split samples are used to assess whether the clinical laboratory is meeting the Laboratory Standardization Panel's goals for accuracy and precision. We used a computer-based Monte Carlo simulation model of split-sample proficiency testing to evaluate the certification program and, in particular, to analyze the effects of reference laboratory bias and imprecision. Results of our simulations indicate that the accuracy of the certification program is strongly influenced by reference laboratory bias and less influenced by reference laboratory imprecision. The certification program is potentially highly accurate, but unless reference laboratory bias is tightly controlled, the number of classification errors may limit its utility. Moreover, the decision limit of the certification program needs to be higher than the Laboratory Standardization Panel's goal (3.5% instead of 3.0%) to ensure that an acceptably high proportion of well-performing clinical laboratories can become certified.

Prognostic factors and survival model for decompensated hepatic cirrhosis

A group of 207 consecutive patients admitted for decompensated liver cirrhosis of different etiologies (alcoholic, HBsAg-associated and cryptogenic), was studied in order to assess the independent long-term (up to 5 years) prognostic value of 13 clinical, biochemical and etiological factors. These were analyzed by the Cox Regression Model using a step-wise backward procedure. The final model included bilirubin (p = 0.003), HBsAg (p = 0.006), encephalopathy (p = 0.010) and a factor comprising urea and albumin (p less than 0.001). The model was validated by a split-sample testing technique and may be used to predict survival in decompensated cirrhosis. A comparison with Child-Pugh's score in terms of survival prediction was carried out and was favorable to our model. We conclude that this model can be useful for predicting short and long-term survival in the three most common types of liver cirrhosis and that the additional overhead to calculate it seems justified in view of the large availability of microcomputers where simple programs can be run to perform this task and draw the predicted survival curves.

Assessment of split-sample proficiency testing for cholesterol by use of a computer simulation model

We developed a computer model to study the use of patients' specimens to assess compliance of cholesterol measurement performance with the 1992 goals of the Laboratory Standardization Panel of the National Cholesterol Education Program. The model uses Monte Carlo techniques to simulate cholesterol measurements that are subject to both systematic and random error. Split-sample measurements by a clinical laboratory and by a reference laboratory are compared by using linear regression to estimate clinical laboratory bias and imprecision; subsequently, according to specified decision limits, the performance of the clinical laboratory is classified as acceptable or deficient. We have quantified the influence of the bias and imprecision of the clinical laboratory, the imprecision of the reference laboratory, the number of split samples compared, and the decision limits on the accuracy of the classification of clinical laboratory performance. Unless the decision limits are carefully selected and a sufficient number of split samples are used, clinical laboratory performance will be frequently misclassified.

Prognosis after the first episode of gastrointestinal bleeding or coma in cirrhosis. Survival and prognostic factors.

Hepatic encephalopathy and gastrointestinal (GI) bleeding are the most serious complications in cirrhosis. The purpose of this study was to examine survival after the first episode of GI bleeding or coma, or both, and to identify variables associated with the subsequent survival in 284 consecutive patients with cirrhosis admitted to one division of hepatology over a period of 81 months. Patients who only bled had markedly longer survival than those who only had coma, whereas those who had both bleeding and coma had by far the poorest survival, only 15% being alive 1 year later. Several other variables showed a significant association with survival. In a Cox multiple regression analysis the following four variables showed significant association with a short survival: coma and bleeding at the episode, ascites, low prothrombin index, and high serum creatinine. The prognostic index derived from the Cox model, which was validated by a split-sample testing technique, may be used to refine prognostic estimation in this subgroup of severely ill cirrhotic patients.